The Effect of Smoking on the Relationship between Alcohol Consumption and All-Cause Mortality in Canadian Women

There is substantial evidence that alcohol consumption can be detrimental to both health and social life (for overviews see Rehm et al., 2004; Babor et al., 2003; Ezzati et al., 2002; Klingemann & Gmel, 2001; United States Department of Health and Human Services, 1997; Edwards et al., 1994). However, alcohol consumption also has health and social benefits (Rehm et al., 2003a; Klingemann & Gmel, 2001; Ashley et al., 2000), most notably the beneficial effect of moderate drinking on coronary heart disease (CHD) (Rehm et al., 2003b). All-cause mortality is an important indicator of the severe health consequences of alcohol consumption (United States Department of Health and Human Services, 2000).The exact shape of the relationship between average amount of alcohol consumption and all-cause mortality depends on the distribution of disease outcomes (i.e., the contribution of cancer or CHD deaths to overall deaths) (Murray & Lopez, 1996) and on patterns of drinking (Rehm et al., 2004; Rehm et al., 2001b), but it can usually be described as J-shaped in both sexes in age groups older than 45 (Rehm & Sempos, 1995a). Thus the most recent meta-analysis described abstainers as having a higher mortality risk than "on average" moderate drinkers, with "on average" heavier drinkers displaying the highest risk for mortality (Corrao et al., 2000).However, there are still important questions concerning the relationship between alcohol consumption and mortality. First, there have been few studies on alcohol consumption in women, resulting in point estimates with relatively wider confidence intervals in meta-analyses, especially for the "on average" heavier drinking categories (Rehm et al., 2001b; Gmel et al., 2003). second, the role of smoking has not been fully clarified. Kozlowski and Ferrence (1990) presented evidence that the J-shaped function may largely be an artefact of inadequate control for smoking status and smoking history. In their analyses, never-smoking alcohol abstainers died at about the same rate as never-smoking moderate drinkers. The authors postulated that the protective effect was confined to current and former smokers. They further stressed the importance of using precise measures of substance use history and status, and they indicated that given the strong covariation between smoking and drinking, usual approaches to the statistical control of smoking-namely, inclusion of smoking as a covariate in the regression equation-may produce misleading results. Unfortunately, most studies either do not have large enough sample sizes to do separate analyses by smoking status or they do not have detailed enough information on smoking (for an exception, see Rehm &Sempos, 1995b).The purpose of this analysis is to help fill the above knowledge gaps with respect to the relationship between alcohol, smoking, and mortality in women. The Canadian National Breast Screening Study (NBSS) cohort in Canada had a large enough sample size to have sufficient power to stratify the analyses by smoking, and thus to detect effects of smoking in heavy drinkers.MethodsSampleThe NBSS was a multi-center controlled clinical trial of (primarily) mammographie screening for breast cancer in 89,835 women ages 40-59 years at recruitment, between 1980 and 1985 (Miller et al., 1981). For this investigation, eligibility was restricted to a subgroup of women who by design received a self-administered quantitative food-frequency questionnaire in addition to the standard epidemiologic questionnaire, given that alcohol consumption was assessed using the food-frequency questionnaire (Jain et al., 1982). The resulting effective sample size of this investigation was 49,165 women. The final sample was obtained after excluding all women with extreme values for total energy intake (N=167) or body mass index (BMI: weight in kg/height in nr) (N=279). The cutoffs for extreme values were lower than 14 and more than 65 kg/m^sup 2^ for BMI, and more than three standard deviations above or below the mean for total energy intake. …